Kidney dysfunction is a common finding in

AMERICAN ASSOCIATION FOR THE STUDY OFLIVERD I S E ASES HEPATOLOGY, VOL. 65, NO. 2, 2017 Development and Validation of a Mathematical Equation to Estimate Glomerular Filtration Rate in Cirrhosis: The Royal Free Hospital Cirrhosis Glomerular Filtration Rate Maria Kalafateli, 1 Fred Wickham, 2 Maria Burniston, 2 Evangelos Cholongitas, 3 Eleni Theocharidou, 1 Matteo Garcovich, 1 James O Beirne, 1 Rachel Westbrook, 1 Gioacchino Leandro, 4 Andrew K. Burroughs, 1 and Emmanuel A. Tsochatzis 1 Current expressions based on serum creatinine concentration overestimate kidney function in cirrhosis, leading to significant differences between true and calculated glomerular filtration rate (GFR). We compared the performance of the four-variable and six-variable Modification of Diet in Renal Disease and chronic kidney disease epidemiology with true, or measured, GFR (mgfr) and the impact of this difference on Model for End-Stage Liver Disease (MELD) calculation. We subsequently developed and validated a GFR equation specifically for cirrhosis and compared the performance of the new derived formula with existing GFR formulae. We included 469 consecutive patients who had a transplant assessment between 2011 and 2014. mgfr was measured using plasma isotope clearance according to a technique validated in patients with ascites. A corrected creatinine was derived from the mgfr after application of the Modification of Diet in Renal Disease formula. Subsequently, a corrected MELD was calculated and compared with the conventionally calculated MELD. Stepwise multiple linear regression was used to derive a GFR equation. This was compared with the mgfr in independent external and internal validation sets of 82 and 174 patients with cirrhosis, respectively. A difference >20 between existing formulae and mgfr was observed in 226 (48.2%) patients. The corrected MELD score was 3 points higher in 177 (37.7%) patients. The predicted equation (r 2 5 74.6%) was GFR 5 45.9 3 (creatinine 0836 ) 3 (urea 0229 ) 3 (international normalized ratio 0113 ) 3 (age 20.129 [Corrected November 29, 2016: originally written as age-129. ]) 3 (sodium 0972 ) 3 0.809 (if female) 3 0.92 (if moderate/severe ascites). An online calculator is available at http://rfh-cirrhosis-gfr.ucl.ac.uk. The model was a good fit and showed the greatest accuracy compared to that of existing formulae. Conclusion: We developed and validated a new accurate model for GFR assessment in cirrhosis, the Royal Free Hospital cirrhosis GFR, using readily available variables; this remains to be tested and incorporated in prognostic scores in patients with cirrhosis. (HEPATOLOGY 2017;65:582-591). Kidney dysfunction is a common finding in patients with cirrhosis and is associated with increased mortality. (1,2) This is reflected by the inclusion of serum creatinine concentration in the Model for End-Stage Liver Disease (MELD) score, which is a prognostic tool used for liver transplant prioritization. However, the use of creatinine for kidney function assessment in patients with cirrhosis can lead Abbreviations: CI, confidence interval; CKD-EPI, chronic kidney disease epidemiology; GFR, glomerular filtration rate; INR, international normalized ratio; IQR, interquartile range; MDRD, Modification of Diet in Renal Disease; MELD, Model for End-Stage Liver Disease; mgfr, measured GFR; OR, odds ratio; RFH, Royal Free Hospital. Received January 31, 2016; accepted September 25, 2016. Additional Supporting Information may be found at onlinelibrary.wiley.com/doi/10.1002/hep.28891/suppinfo. Supported by the Hellenic Association for the Study of the Liver (to M.K.). Copyright VC 2016 by the American Association for the Study of Liver Diseases View this article online at wileyonlinelibrary.com. DOI 10.1002/hep.28891 Potential conflict of interest: Nothing to report. 582

HEPATOLOGY, Vol. 65, No. 2, 2017 KALAFATELI ET AL. to systematic bias. Decreased creatinine production, increased tubular creatinine excretion, muscle depletion, and the interference of high bilirubin levels with the analytic methods used for determination of creatinine (especially with the Jaffe reaction) may contribute to falsely low serum creatinine levels, thus leading to an overestimation of kidney function and an underestimation of liver disease severity using the MELD score. (3) Accurate measurement of glomerular filtration rate (GFR) is particularly challenging in patients with liver disease, and there is not a clear consensus as to which technique is the best reference standard. Although plasma clearance methods for assessment of GFR are recommended, because of technical difficulties, lack of availability, and high cost, these are not readily available to use in routine clinical practice. Furthermore, they are prone to inaccuracy, particularly in patients with fluid retention such as ascites. (4) The current British Nuclear Medicine Society guidelines for the measurement of GFR using plasma sampling recommend not using plasma clearance assessment of GFR in patients with ascites, edema, or other expanded body space. (5) Recently, however, Wickham et al. have described a modified plasma clearance method for assessment of GFR that can be used in liver patients with ascites. (6,7) For liver transplant candidates without ascites, this method showed good agreement with the slope intercept technique described in the current guidelines with plasma samples taken at 2, 4, and 6 hours postinjection. The creatinine-based equations used for estimation of GFR, which did not include patients with cirrhosis when first developed, are poor predictors of kidney function in patients with cirrhosis, leading to a >20% overestimation of true, or measured, GFR (mgfr). (8,9) The primary objective of this study was to develop and validate a GFR equation specifically for patients with cirrhosis and to compare the performance of the new derived formula with the existing GFR formulae of four-variable and six-variable Modification of Diet in Renal Disease (MDRD) and chronic kidney disease epidemiology (CKD-EPI). Secondary objectives were to assess the differences between mgfr and estimated GFR using common GFR formulae and between observed and corrected MELD score and to evaluate the predictors of these differences. Patients and Methods From January 2011 to September 2014, 469 consecutive patients with cirrhosis evaluated for liver transplantation at the Royal Free Hospital (RFH) were included in the study and comprised the training data set. An independent cohort of consecutive patients with cirrhosis (n 5 82) with available chromium-ethylene diamine tetraacetic acid and cystatin measurements that were evaluated at the Hippokration General Hospital of Thessaloniki in Greece was included as an external validation set. The internal validation set included 174 patients with cirrhosis assessed for a liver transplant between February 2007 and December 2010. The GFR measurement using isotope plasma clearance was part of the patients standard pretransplant workup. Patients with simultaneous multiple organ transplantation, acute liver failure, or prior liver transplantation were excluded from the analysis. The chair of the Hampstead Research Ethics Committee was consulted and agreed that, as all measurements made were in accordance with clinical guidelines and only additional calculations were involved, the study raised no ethical issues, and patient consent was not required. The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki. ARTICLE INFORMATION: From the 1 UCL Institute for Liver and Digestive Health and Sheila Sherlock Liver Unit, Royal Free Hospital and University College London, London; 2 Department of Nuclear Medicine, Royal Free Hospital, London, UK; 3 4th Department of Internal Medicine, Hippokration General Hospital of Thessaloniki, Medical School of Aristotle University, Thessaloniki, Greece; 4 National Institute of Gastroenterology, S. de Bellis Research Hospital, Castellana Grotte, Italy. ADDRESS CORRESPONDENCE AND REPRINT REQUESTS TO: Emmanuel A. Tsochatzis UCL Institute for Liver and Digestive Health, Royal Free Hospital and University College London London, UK E-mail: e.tsochatzis@ucl.ac.uk Tel: 144-2077940500, ext. 31142 583

KALAFATELI ET AL. HEPATOLOGY, February 2017 CREATININE ASSESSMENT Creatinine concentration was measured in each sample using the O Leary modified Jaffe method, which shows the least interference with bilirubin levels. (10) Creatinine measurements were made on a Roche analyzer using Roche reagents in the RFH (Roche Diagnostics GmbH). In the external validation cohort, compensated Jaffe was used for creatinine measurements (Beckman Coulter) and cystatin C was analyzed by immunonephelometry using a BN- ProSpec analyzer (Dade Behring BN-ProSpec; reference range 0.53-0.95 mg/l). All creatinine measurements were standardized to the Information Display Measurements Standard. GFR ASSESSMENT USING RADIOISOTOPE PLASMA CLEARANCE For patients in the training and internal validation data sets, assessment of GFR using radioisotope plasma clearance was carried out as described. (6,7) For patients in the external validation set, assessment of GFR was performed as described. (11) Full details are provided in the Supporting Information. FORMULAE FOR GFR ESTIMATION All formulae were calculated according to published data. (12-14) Details are provided in the Supporting Information. STATISTICAL ANALYSIS Descriptive Analysis The MDRD study equation was rearranged to give an expression for creatinine concentration in terms of GFR. This expression was used to calculate a value for corrected creatinine concentration from mgfr. Subsequently, a corrected MELD was calculated. The Wilcoxon signed rank test was used to assess the differences between corrected and observed MELD scores. Full details are given in the Supporting Information. Derivation and Validation of the New Equation for GFR Estimation Backward stepwise multiple linear regression on log-transformed data was used to derive a new GFR equation using the training set. The SPSS default P values for removing or reentering variables were used. Data were logarithmically transformed to eliminate the great variance across the range of GFR and subsequently reexpressed in their original units. (12,15) The following variables were considered in the univariate analysis: age, sex, ethnicity, mean arterial pressure, dry weight, height, body mass index, hand grip strength, international normalized ratio (INR), serum albumin, urea, creatinine, total bilirubin and sodium levels, presence and severity of ascites and encephalopathy, etiology of liver disease, and MELD and Child-Pugh scores. (12,15) The regression coefficients determined in the training set were applied to obtain the predicted GFRs in the validation set. To determine if the new equation fits the data well, we calculated the r 2 statistic, the mean difference between observed and predicted GFR (residual) values in the validation set, the root mean square error (standard deviation of the mean difference), and the appropriate residual plots. Performance and Comparison of the Different Equations to Predict GFR in the Training and Validation Set Bias was assessed as the median difference between mgfr and estimated GFR using the new equation as well as MDRD and CKD-EPI study equations, with negative values indicating an overestimation of mgfr. (12) Precision was assessed as interquartile range (IQR) for the differences. (12) Accuracy was assessed as the percentage of predictions within 10% (P10), 30% (P30), and 50% (P50) of mgfr. (8) Confidence intervals of median difference, IQR, and P10, P30, and P50 were estimated with the bootstrap method (200 bootstraps). Significance testing was twosided and set to <0.05. Analysis was performed using the SPSS statistical package (version 22.0; IBM, New York, NY) and MedCalc for Windows (version 12.5; MedCalc Software, Ostend, Belgium). Results DEMOGRAPHIC AND CLINICAL CHARACTERISTICS The baseline patients characteristics in the training and validation cohorts are shown in Table 1. There were significant differences in urea, albumin, sodium, 584

HEPATOLOGY, Vol. 65, No. 2, 2017 KALAFATELI ET AL. TABLE 1. Baseline Characteristics of the Study Population Training set (n 5 469) Internal validation set (n 5 174) External validation set (n 5 82) P1* P2 Age (years), median (range) 55 (16-76) 54 (21-71) 53 (16-75) 0.540 0.551 Sex (M/F), n (%) 324/145 (69.1/30.9) 117/57 (67.2/32.8) 60/22 (73.2/26.8) 0.411 0.457 Black, n (%) 29 (6.2) 9 (5.2) 0 0.471 0.021 Etiology of liver disease, n (%) Alcohol 142 (30.3) 37 (21.3) 24 (29.3) 0.137 0.073 Viral hepatitis 135 (28.8) 56 (32.2) 36 (43.9) PSC 51 (10.9) 25 (14.4) 9 (11) PBC 25 (5.3) 11 (6.3) 3 (3.7) NASH/cryptogenic 54 (11.5) 13 (7.4) 10 (12.2) Alcohol and viral 4 (0.9) 15 (8.6) 0 Other 32 (6.8) 15 (14.1) 0 HCC, n (%) 111 (23.7) 54 (31) 8 (9.8%) 0.070 0.005 BMI (kg/m 2 ), median (range) 26.9 (16.7-49.1) 26.5 (16.9-47.7) 25.6 (19.3-46.1) 0.725 0.195 BSA (m 2 ), mean 6 SD 1.9 6 0.3 1.9 6 0.2 1.9 6 0.2 0.819 0.171 HGS (kg), median (range) 24 (2-47) 26.5 (9.5-56) NA 0.060 NA MAP (mm Hg), median (range) 83 (59-122) 83.3 (63-103) NA 0.197 NA INR, median (range) 1.3 (0.9-5.8) 1.4 (0.9-2.8) 1.3 (1.0-7.2) 0.429 0.686 Albumin (mg/dl), median (range) 33 (16-49) 35 (18-52) 31 (16-45) 0.002 0.094 Bilirubin (lmol/l), median (range) 37 (3-639) 36 (4-932) 33.1 (7-397) 0.675 0.720 Sodium (mmol/l), median (range) 138 (113-147) 139 (120-149) 137 (117-145) 0.023 0.346 Urea (mmol/l), median (range) 5.2 (1.4-31.1) 4.8 (1.8-26.2) 5.7 (2.2-36.4) 0.029 0.132 Creatinine (lmol/l), median (range) 75 (33-464) 72 (39-261) 85.8 (43-261) 0.135 0.002 Corrected creatinine (lmol/l), median (range) 81.06 (11.1-162.5) NA NA 0.251 NA mgfr ( ), median (range) 66 (7-129) 78.1 (5.7-141) 73 (16-150) <0.001 0.048 GFR MDRD-4 ( ), median (range) 87.04 (9.7-230.6) 90.9 (18.9-230.8) 76.55 (23.08-168.47) 0.359 0.001 GFR MDRD-6 (ml/minute/1.73 m 2 ), median (range) 81.63 (10.1-193.9) 85.2 (19.4-178.1) NA 0.57 NA CKD-EPI ( ), median (range) 93.6 (10.3-158.7) 95.0 (20.4-156.4) 85.0 (24.0-145.0) 0.203 0.002 MELD score, median (range) 11 (6-44) 14 (6-37) 13 (5-48) 0.008 0.209 CP score, median (range) 8 (5-13) 7 (5-15) 8 (5-13) <0.001 0.932 CP class, n (%) A 115 (24.5) 65 (37.4) 19 (23.2) 0.004 0.932 B 242 (51.6) 75 (43.1) 42 (51.2) C 112 (23.9) 34 (19.5) 21 (25.6) Ascites, n (%) No/mild 167 (35.6) 122 (70.1) 5 (6.1) <0.001 <0.001 Moderate 182 (38.8) 31 (17.8) 47 (57.3) Severe 120 (25.6) 21 (12.1) 30 (36.6) Encephalopathy, n (%) No 388 (82.7) 140 (80.5) 64 (78.0) 0.046 0.456 Mild/moderate 79 (16.8) 29 (16.7) 18 (22.0) Severe 2 (0.4) 5 (2.9) 0 *Training versus internal validation group. Training versus external validation group Abbreviations: BMI, body mass index; BSA, body surface area; CP, Child-Pugh; HCC, hepatocellular carcinoma; HGS, hand grip strength; MAP, mean arterial pressure; NA, not available; NASH, nonalcoholic steatohepatitis; NS, not significant; PBC, primary biliary cirrhosis; PSC, primary sclerosing cholangitis; SD, standard deviation. and mgfr levels, as well as in the prevalence of ascites and encephalopathy. DIFFERENCE BETWEEN mgfr AND ESTIMATED GFR For the training data set the median difference between estimated GFR and mgfr was 19.1 (IQR 24.1) and 19.9 (IQR 22.7) for MDRD and CKD-EPI formulae, respectively. Plots of estimated GFR versus the difference between mgfr and estimated GFR showed a consistent overestimation of GFR when using these formulae (Fig. 1A-C; Supporting Fig. S1A,B). A difference >20 between MDRD and mgfr was observed in 226 (48.2%) patients. In multivariate binary logistic regression analysis, this difference was independently associated with male sex (odds ratio [OR] 5 3.5, 95% confidence interval [CI] 2.12-5.8), moderate/severe ascites (OR 5 585

KALAFATELI ET AL. HEPATOLOGY, February 2017 2.1, 95% CI 1.31-3.38), and serum levels of sodium (OR 5 0919, 95% CI 0.873-0.967), creatinine (OR 5 0.949, 95% CI 0.937-0.961), and bilirubin (OR 5 1.003, 95% CI 1.001-1.005). The difference was more pronounced in patients with worsening Child-Pugh class (Supporting Table S1). MELD SCORE USING CORRECTED CREATININE LEVELS The median corrected MELD was 14 (range 6-37) and significantly higher than the observed median MELD score (11, range 6-61) (Wilcoxon test z 5 7.33, P < 0.001). The corrected MELD score was 3 points higher in 177 (37.7%) patients. In the multivariate binary regression analysis, the factors significantly associated with a difference 3 points were high Child-Pugh score (OR 5 1.528, 95% CI 1.367-1.708) and low creatinine levels (OR 5 0.98, 95% CI 0.97-0.988). The proportion of patients with a 3 difference increased along with the severity of liver disease (Child-Pugh A versus B versus C: 19 [16.5%] versus 88 [36.4%] versus 70 [62.5%], respectively; P < 0.001). PREDICTION OF GFR FROM STEPWISE REGRESSION ANALYSIS The variables that were finally included in the multivariate stepwise regression model to estimate log mgfr were log-transformed serum creatinine, urea, INR, age, sodium, bilirubin, albumin, body mass index, ethnicity, sex, mild/moderate/severe encephalopathy, and moderate/severe ascites (Table 2). The derived equation with the maximal r 2 (74.6%) was TABLE 2. Multiple Regression Model to Predict GFR on the Logarithmic Scale Variables Coefficients 95% CI FIG. 1. Plots of estimated GFR versus the difference between mgfr and estimated GFR for (A) MDRD-4, (B) MDRD-6, and (C) CKD-EPI in the validation data set. egfr, estimated GFR. Quantitative Log creatinine (lmol/l) 0.836 0.920 to 0.750 Log urea (mmol/l) 0.229 0.293 to 0.165 Log INR 0.113 0.200 to 0.023 Log age (years) 0.129 0.217 to 0.042 Log sodium (mmol/l) 0.972 0.320-1.620 Qualitative Sex (female) 0.092 0.113 to 0.072 Moderate/severe ascites 0.0369 0.058 to 0.015 586

HEPATOLOGY, Vol. 65, No. 2, 2017 KALAFATELI ET AL. GFR545:9 3 creatinine 20:836 lmol L 20:229 mmol 3 urea 3 INR 20:113 L 3 age 20:129 ½yearsŠ 3 sodium 0:972 mmol L 3 0:809 ðif femaleþ 3 0:92 if moderate ascites ml=min=1:73 m 2 severe We have named this equation the RFH cirrhosis GFR. Subsequently, this model was applied to the internal validation set to obtain predicted GFR values. The mean difference between observed and predicted GFR (residual) values in the validation set was 3.8 ml/ minute/1.73 m 2, with a root mean square error of 14.8. According to the residual plots, the model was a good fit (Fig. 2; Supporting Fig. S2). PERFORMANCE OF THE NEW EQUATION, MDRD, COCKROFT- GAULT, AND CKD-EPI FORMULAE Table 3 shows the performance of RFH cirrhosis GFR compared with the performances of the fourvariable and six-variable MDRD and CKD-EPI to predict mgfr in both the external and internal validation data sets. The new equation showed the highest performance, followed by the MDRD and CKD-EPI study equations. Plot analysis in the validation groups showed that the new model had greater accuracy than all of the other formulae to predict mgfr (Figs. 2 and 3; Supporting Fig. S1D). To further evaluate the accuracy of the new equation, we calculated the percentage of the predicted GFR with the different formulae within the 10%, 30%, and 50% of mgfr (P10, P30, and P50) in both cohorts. The new equation had the highest accuracy, with P10, P30, and P50 values of 56.1%, 89%, and 98.8% in the external validation cohort and 45.4%, 88.5%, and 96.6% in the internal validation cohort, respectively (Table 3). The performance of the new equation was not influenced by either mgfr or degree of liver dysfunction (Supporting Tables S1 and S2). Although the RFH cirrhosis GFR had better overall accuracy when compared with the cystatin C based equations, this result should be FIG. 2. In the scatter/dot plot of estimated GFR versus the difference between mgfr and estimated GFR for RFH cirrhosis GFR in the internal validation plot; the residuals appear to be randomly scattered about zero. The plots show that the model fits the data well. egfr, estimated GFR. interpreted with caution as the cystatin C assay was not traceable to International Federation of Clinical Chemistry and Laboratory Medicine standards, which could explain the observed overestimate. Discussion We developed and validated the RFH cirrhosis GFR to predict GFR in patients with cirrhosis using data from consecutive patients awaiting liver transplantation. This equation was validated in independent cohorts of patients and showed higher accuracy and less bias than the existing GFR formulae. This is of critical importance because of the shortcomings of the equations based on serum creatinine concentration, which are currently used for estimating kidney function in patients with cirrhosis. (16) Levey et al. (12,15) used urinary clearance of iothalamate for reference measurements of GFR in the derivation of the four-variable MDRD and the CKD-EPI study equations, but Kwong et al. (17) highlighted the limitations associated with the use of this technique caused by inaccuracies in measurement of urine volumes and times, iothalamate concentration, incomplete bladder emptying, and physiologic day-to-day and diurnal fluctuation in GFR. All of these emphasize the importance of the new GFR equation, which was 587

KALAFATELI ET AL. HEPATOLOGY, February 2017 TABLE 3. Comparison of the Performances of the New Equation, MDRD, and CKD-EPI Equations in the Training and External and Internal Validation Cohorts New equation MDRD-4 MDRD-6 CKD-EPI CKD-EPI cystatin C CKD-EPI cystatin C-creatinine Training cohort Median difference (95% CI), 19.2 ( 21.3 to 17.1) 13.0 ( 15.1 to 11.4) 19.9 ( 22.4 to 17.9) NA NA IQR for differences (95% CI), 24.1 (21.7-27.3) 21.3 (18.2-23.6) 22.65 (19.9-25.6) NA NA P10 (%) (95% CI) 19 (15.3-22.4) 24.1 (20.6-28.2) 17.7 (14.5-21.1) NA NA P30 (%) (95% CI) 47.8 (43.1-52.2) 60.1 (56.1-64.9) 45 (40.5-49.5) NA NA P50 (%) (95% CI) 69.1 (65.3-73.3) 80.8 (77.6-84.4) 67.8 (63.5-72.1) NA NA External validation cohort Median difference (95% CI), 6.5 (4.9, 8.4) 4.7 ( 9.1 to 0.5) 22.6 ( 27.4 to 16.2) 7.0 ( 13.0 to 2.1) 31.0 (24.4-35.0) 16.5 (13.8-22.0) IQR for differences (95% CI), 10.9 (7.9-13.3) 23.0 (17.2-30.2) 27.1 (22.5-33.4) 20.0 (17.3-27.7) 25.0 (19.0-35.5) 17.5 (14.3-28.1) P10 (%) (95% CI) 56.1 (42.2-68.3) 29.3 (18.3-39.5) 19.5 (13-30) 31.7 (20.3-44.5) 9.8 (3.7-14.6) 15.9 (8.0-25.1) P30 (%) (95% CI) 89.0 (80.7-95.7) 75.6 (66.9-86.0 46.3 (36.6-57.6) 75.6 (65.3-84.1) 26.8 (17.8-37.1) 65.9 (56.1-75.6) P50 (%) (95% CI) 98.8 (95.1-100.0) 92.7 (86.0-97.6) 76.8 (66.7-87.9) 89.0 (80.1-95.0) 64.6 (53.7-74.9) 95.1 (89.0-98.8) Internal validation cohort Median difference (95% CI), 3.0 (0.4-6.4) 14.2 ( 17.6 to 10.4) 10.6 ( 13.1 to 9.1) 14.9 ( 19.0 to 11.7) NA NA IQR for differences (95% CI), 20.4 (16.7-24.9) 20.8 (17.6-26.1) 19.0 (15.6-23.2) 21.2 (17.2-24.2) NA NA P10 (%) (95% CI) 45.4 (37.0-54.0) 25.9 (19.9-32.5) 28.2 (22.6-35.1) 24.1 (17.4-31.4) NA NA P30 (%) (95% CI) 88.5 (83.1-93.2) 60.9 (53.8-69.2) 70.7 (62.6-77.6) 59.2 (51.5-66.5) NA NA P50 (%) (95% CI) 96.6 (93.4-98.9) 81.0 (75.0-88.0) 85.1 (79.0-89.8) 77.0 (69.9-83.6) NA NA Abbreviation: NA, not available. 588

HEPATOLOGY, Vol. 65, No. 2, 2017 KALAFATELI ET AL. FIG. 3. Plot of estimated GFR versus the difference between mgfr and estimated GFR for RFH cirrhosis GFR in the external validation data set. egfr, estimated GFR. specifically developed and validated in patients with cirrhosis and takes into consideration potential variables that affect kidney function in this setting, together with known predictors of GFR such as age and sex. The MDRD and CKD-EPI study equations systematically overestimated kidney function, and this was more pronounced in patients with worse liver function. Indicatively, the difference between MDRD and the true GFR was >20 in approximately 50% of patients with cirrhosis. A similar discrepancy was observed using the CKD-EPI expressions (data not shown). As shown in the multivariate analysis, patients with higher bilirubin levels and lower serum sodium, and thus more impaired liver function, were more likely to have an overestimation of kidney function using the MDRD formula. Male sex was also independently associated with such a difference; therefore, male patients are disadvantaged using the MDRD formula, implying that the regression coefficient used for gender in the MDRD equation is inappropriate when used in cirrhosis. Presence of ascites also resulted to higher rates of overestimating true GFR; this is in line with the results by Francoz et al., (9) who showed that 46% of 157 patients with cirrhosis had a GFR overestimation of 20% with the MDRD study equation. Finally, patients with lower creatinine levels are more likely to have an overestimation of GFR when using the MDRD study equation. This reflects the already discussed discrepancies between measured creatinine and renal function in cirrhosis that are more profound in patients with creatinine concentrations within the normal range. (9) Therefore, a number of patients with impaired kidney function (low GFR) and, thus, at a higher mortality risk but with low serum creatinine are significantly overscored with the existing equations for GFR calculation. The inaccuracy of creatinine at predicting kidney function and, subsequently, mortality in cirrhosis is reflected by the large proportion of patients (40%) with a 3-point difference between the observed and corrected MELD scores. This has major implications in the current liver transplant allocation system, with some patients being systematically underscored, especially those with advanced liver disease and lower measured creatinine levels. Cholongitas et al. (18) showed that this difference was more profound in female candidates for liver transplantation, suggesting a 3-point correction factor in females with MELD score higher than 19. Gender and age, which are considered significant determinants of serum creatinine (19) as a result of their correlation with muscle mass, were independent predictors of GFR and were included in the new equation, similar to existing GFR formulae. Patients with cirrhosis are commonly malnourished and/or sarcopenic, (20) and thus, the reduced muscle mass has a different impact on creatinine generation from that observed in the general population. Female patients have lower creatinine levels for the same GFR values compared to male patients with cirrhosis (18) ; therefore, gender was included in the new equation but with a different weight from the one used for GFR estimation in the other creatinine-based formulae. Urea was also an independent predictor of GFR, reflecting the correlation between urea clearance and GFR; there is a difference in the ratios of the amount secreted by the tubule and the amount filtered by the glomerulus between urea and creatinine and, thus, although both are determinants of renal function, their serum levels fluctuate independently. In cirrhosis, circulating blood urea nitrogen might be increased as a result of occult gastrointestinal bleeding due to portal hypertensive gastropathy or use of steroids. Race did not have an impact on kidney function in patients with cirrhosis. Although this might be a type 2 error due to the low number of patients of black ancestry in our cohort, it seems that the weight given to race by the other creatinine-based equations is unsuitable for patients with cirrhosis compared to the general population or to patients with 589

KALAFATELI ET AL. HEPATOLOGY, February 2017 chronic kidney insufficiency. On the other hand, the presence of moderate/severe ascites together with high INR values and low sodium levels had a negative correlation with GFR estimation, reflecting the impaired kidney function that accompanies patients with largevolume ascites and/or advanced liver disease (21,22) due to the chronically reduced renal blood flow. (23) The accuracy of the new equation was satisfactory and significantly better than the accuracy of existing formulae (89% versus 27%-75% of estimates being within 30% of true GFR, respectively). The accuracy of existing formulae in our cohort is in agreement with previous reports (8) that showed that only 60%-66% of estimates (using Cockroft-Gault; MDRD four, five, and six variables; and the Nankivell formulae) were within 30% of the mgfr in a large cohort of 1,147 patients with cirrhosis. Precision, measured by the IQR for the differences, did not differ among the currently used formulae and is similar to that reported by Levey et al. in the CKD-EPI prediction equation study. (12) The new equation, apart from being significantly more accurate, has several other advantages over the existing equations in patients with liver disease. Firstly, it was derived from a cohort of patients with cirrhosis at various stages of disease severity (MELD ranging 6-44), including patients with hepatocellular carcinoma, and has been further validated in an independent cohort comprised of patients with diverse clinical characteristics. Secondly, it can be easily implemented in routine clinical practice as it includes readily available variables. Thirdly, it does not include variables such as albumin and bilirubin that can be influenced by several factors such as albumin infusions or that require calibration in different laboratories for optimal use. (24) Lastly, it predicts GFR across a wide range of values, allowing general applicability in patients with cirrhosis. The main limitation of the new equation is the use of creatinine as the major determinant of GFR, which is influenced by several factors unrelated to kidney function; however, creatinine is the most readily available predictor of renal function, and we tried to eliminate its weaknesses by including in the model other extrarenal determinants of renal function such age, gender, and liver disease severity. The equations based on cystatin C, a protein that is eliminated almost exclusively by glomerular filtration, have better performance than the creatinine-based equations in patients with cirrhosis. (25) However, the use of cystatin C is not without limitations including the high cost, its interference with several drugs such as steroids, the lack of standardization, and its unsuitability in infectious conditions which are very common in end-stage liver disease. (16) In an exploratory analysis, the RFH cirrhosis GFR equation performed better than cystatin C based equations in the external validation cohort, but this needs to be further validated in larger series. Secondly, the prediction equation has been applied in a small proportion of black patients (6%) and only in the pretransplant setting; therefore, this will need further validation in non-caucasian populations. Finally, although the sample sizes of the training and validation cohorts are relatively small compared to other validation studies of estimating equations in the general population, (12,15) they are considerably larger than those in previous studies evaluating kidney function in cirrhosis; and the findings were quite robust. (26) In conclusion, inaccurate estimation of kidney function has major implications both in terms of prognosis and in terms of candidate selection and prioritization for liver transplantation. We therefore developed and validated an accurate cirrhosis-specific equation for indirect GFR assessment in patients with varying disease severity, taking into consideration the most important renal and extrarenal determinants of renal function in cirrhosis. An online calculator is available at http://rfh-cirrhosisgfr.ucl.ac.uk. The RFH cirrhosis GFR performs significantly better than existing equations such as MDRD and CKD-EPI. We strongly encourage other research teams to independently validate the performance of this equation in larger populations of patients with cirrhosis and diverse clinical characteristics. The incorporation of this cirrhosis-specific GFR equation instead of creatinine in prognostic scores such as MELD should be further tested as it is highly likely that it will increase their overall performance. REFERENCES 1) Fede G, D Amico G, Arvaniti V, Tsochatzis E, Germani G, Georgiadis D, et al. Renal failure and cirrhosis: a systematic review of mortality and prognosis. J Hepatol 2012;56:810-818. 2) Tsochatzis EA, Bosch J, Burroughs AK. Liver cirrhosis. Lancet 2014;383:1749-1761. 3) Cholongitas E, Shusang V, Marelli L, Nair D, Thomas M, Patch D, et al. Review article: renal function assessment in cirrhosis difficulties and alternative measurements. Aliment Pharmacol Ther 2007;26:969-978. 4) Henriksen UL, Henriksen JH. The clearance concept with special reference to determination of glomerular filtration rate in patients with fluid retention. Clin Physiol Funct Imaging 2015; 35:7-16. 5) Fleming JS, Zivanovic MA, Blake GM, Burniston M, Cosgriff PS; British Nuclear Medicine Society. Guidelines for the 590

HEPATOLOGY, Vol. 65, No. 2, 2017 KALAFATELI ET AL. measurement of glomerular filtration rate using plasma sampling. Nucl Med Commun 2004;25:759-769. 6) Wickham F, Burniston MT, McMeekin H, Hilson AJ, Burroughs AK. Validation and impact of a new technique for assessment of glomerular filtration rate in patients with liver disease. Nucl Med Commun 2015;36:168-179. 7) Wickham F, Burniston MT, Xirouchakis E, Theocharidou E, Wesolowski CA, Hilson AJ, et al. Development of a modified sampling and calculation method for isotope plasma clearance assessment of the glomerular filtration rate in patients with cirrhosis and ascites. Nucl Med Commun 2013;34:1124-1132. 8) Gonwa TA, Jennings L, Mai ML, Stark PC, Levey AS, Klintmalm GB. Estimation of glomerular filtration rates before and after orthotopic liver transplantation: evaluation of current equations. Liver Transpl 2004;10:301-309. 9) Francoz C, Prie D, Abdelrazek W, Moreau R, Mandot A, Belghiti J, et al. Inaccuracies of creatinine and creatinine-based equations in candidates for liver transplantation with low creatinine: impact on the model for end-stage liver disease score. Liver Transpl 2010;16:1169-1177. 10) Cholongitas E, Marelli L, Kerry A, Senzolo M, Goodier DW, Nair D, et al. Different methods of creatinine measurement significantly affect MELD scores. Liver Transpl 2007;13:523-529. 11) Cholongitas E, Ioannidou M, Goulis I, Chalevas P, Ntogramatzi F, Athanasiadou Z, et al. Comparison of creatinine and cystatin formulae with 51 chromium-edta glomerular filtration rate in patients with decompensated cirrhosis. J Gastroenterol Hepatol 2016; doi:10.1111/jgh.13446. 12) Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF 3rd, Feldman HI, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med 2009;150:604-612. 13) Wiesner R, Edwards E, Freeman R, Harper A, Kim R, Kamath P, et al. Model for End-Stage Liver Disease (MELD) and allocation of donor livers. Gastroenterology 2003;124:91-96. 14) Du Bois D, Du Bois EF. A formula to estimate the approximate surface area if height and weight be known. Arch Intern Med 1916;17:863-871. 15) Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group. Ann Intern Med 1999;130: 461-470. 16) Francoz C, Glotz D, Moreau R, Durand F. The evaluation of renal function and disease in patients with cirrhosis. J Hepatol 2010;52:605-613. 17) Kwong YT, Stevens LA, Selvin E, Zhang YL, Greene T, Van Lente F, et al. Imprecision of urinary iothalamate clearance as a gold-standard measure of GFR decreases the diagnostic accuracy of kidney function estimating equations. Am J Kidney Dis 2010; 56:39-49. 18) Cholongitas E, Marelli L, Kerry A, Goodier DW, Nair D, Thomas M, et al. Female liver transplant recipients with the same GFR as male recipients have lower MELD scores a systematic bias. Am J Transplant 2007;7:685-692. 19) Perrone RD, Madias NE, Levey AS. Serum creatinine as an index of renal function: new insights into old concepts. Clin Chem 1992;38:1933-1953. 20) Cheung K, Lee SS, Raman M. Prevalence and mechanisms of malnutrition in patients with advanced liver disease, and nutrition management strategies. Clin Gastroenterol Hepatol 2012;10:117-125. 21) Fernandez-Esparrach G, Sanchez-Fueyo A, Gines P, Uriz J, Quinto L, Ventura PJ, et al. A prognostic model for predicting survival in cirrhosis with ascites. J Hepatol 2001;34:46-52. 22) Salerno F, Borroni G, Moser P, Badalamenti S, Cassara L, Maggi A, et al. Survival and prognostic factors of cirrhotic patients with ascites: a study of 134 outpatients. Am J Gastroenterol 1993;88:514-519. 23) Rivolta R, Maggi A, Cazzaniga M, Castagnone D, Panzeri A, Solenghi D, et al. Reduction of renal cortical blood flow assessed by Doppler in cirrhotic patients with refractory ascites. HEPATOL- OGY 1998;28:1235-1240. 24) Levey AS, Coresh J, Greene T, Stevens LA, Zhang YL, Hendriksen S, et al. Using standardized serum creatinine values in the Modification of Diet in Renal Disease study equation for estimating glomerular filtration rate. Ann Intern Med 2006;145: 247-254. 25) De Souza V, Hadj-Aissa A, Dolomanova O, Rabilloud M, Rognant N, Lemoine S, et al. Creatinine- versus cystatine C-based equations in assessing the renal function of candidates for liver transplantation with cirrhosis. HEPATOLOGY 2014;59:1522-1531. 26) Mindikoglu AL, Dowling TC, Weir MR, Seliger SL, Christenson RH, Magder LS. Performance of chronic kidney disease epidemiology collaboration creatinine-cystatin C equation for estimating kidney function in cirrhosis. HEPATOLOGY 2014; 59:1532-1542. Supporting Information Additional Supporting Information may be found at onlinelibrary.wiley.com/doi/10.1002/hep.28891/suppinfo. 591